Re: [PATCH v3 19/36] thunderbolt: Extend tunnel creation to more than 2 adjacent switches

[Mika Westerberg]

On Mon, Apr 08, 2019 at 11:49:04AM +0200, Lukas Wunner wrote:
> On Mon, Apr 08, 2019 at 12:07:44PM +0300, Mika Westerberg wrote:
> > On Mon, Apr 08, 2019 at 10:53:37AM +0200, Lukas Wunner wrote:
> > > Hm, what other cases are there, i.e. what is the meaning of a tb_regs_hop's
> > > "next_hop" field if "out_port" doesn't have a remote?  (And why does it
> > > need to be tracked on the out_port?  In case a remote is added later?)
> > 
> > We also need to program HopIDs for adapter ports (PCIe, DP, NHI) in
> > order to enable a path. The "next_hop" from NULL port to an adapter port
> > tells the HopID a packet gets when it is routed to the adapter port and
> > the adapter port registers then are used to specify which HopID means
> > what (for PCIe there is only 8 but for DP there is 8 and 9, for NHI it
> > can be anything the service driver has negotiated).
> 
> Okay, so in_hopids are the entries allocated in this port's hop config
> space, whereas out_hopids only really bears significance for adapter
> ports.  For null ports (in-between two adapter ports on a path),
> out_hopids is identical to the in_hopids of the next hop (IIUC).
> 
> That probably merits mentioning in struct tb_port's kerneldoc.

Sure, will do.